I am happy to share our paper with BHK LAB at Princess Margaret Cancer Centre!

We studied generalization of drug response predictors in different scenarios across multiple drugs and datasets. IC50 is the most widely-used measure of drug sensitivity and GDSCv1 is the dataset currently most often used for training drug response predictors but our experimental results showed that AAC is a better metric to train machine learning models. We also investigate the impact of hematopoietic tissues on training machine learning models. It is important to remove samples of non-solid tissues from the training dataset.